Rebooting into a new kernel with kexec fails (system dies) if tried on
a machine that has no-execute support. Reason for this is that the so
called datamover code gets executed with DAT on (MMU is active) and
the page that contains the datamover is marked as non-executable.
Therefore when branching into the datamover an unexpected program
check happens and afterwards the machine is dead.
This can be simply avoided by disabling DAT, which also disables any
no-execute checks, just before the datamover gets executed.
In fact the first thing done by the datamover is to disable DAT. The
code in the datamover that disables DAT can be removed as well.
Thanks to Michael Holzheu and Gerald Schaefer for tracking this down.
Reviewed-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Reviewed-by: Philipp Rudo <prudo@linux.vnet.ibm.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Fixes: 57d7f939e7bd ("s390: add no-execute support")
Cc: <stable@vger.kernel.org> # v4.11+
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
s390_reset_system();
data_mover = (relocate_kernel_t) page_to_phys(image->control_code_page);
+ __arch_local_irq_stnsm(0xfb); /* disable DAT - avoid no-execute */
/* Call the moving routine */
(*data_mover)(&image->head, image->start);
ENTRY(relocate_kernel)
basr %r13,0 # base address
.base:
- stnsm sys_msk-.base(%r13),0xfb # disable DAT
stctg %c0,%c15,ctlregs-.base(%r13)
stmg %r0,%r15,gprregs-.base(%r13)
lghi %r0,3
.align 8
load_psw:
.long 0x00080000,0x80000000
- sys_msk:
- .quad 0
ctlregs:
.rept 16
.quad 0